基于ResNet与Attention模块的睡眠呼吸暂停检测分析

Analysis of Sleep Apnea Detection Based on ResNet and Attention Module

摘要: 本文提出了一种基于ResNet架构的模型，用于睡眠呼吸暂停(SAS)的自动检测。该模型以原始心电图(ECG)信号为输入，利用多层卷积和池化结构实现高效的特征提取，省去了额外的特征工程步骤。为了进一步提升模型的性能，特别引入了注意力机制，使模型能够聚焦于信号中关键特征，有效提高了检测的准确性和鲁棒性。实验结果显示，该模型在准确度(93.84%)、灵敏度(92.00%)、特异性(94.97%)及F1分数(0.9193)等指标上均表现优异，显著优于传统网络模型。与近年来相关研究相比，本模型在SAS检测任务中展现了显著优势，突显了其在实际应用中的潜力。

Abstract: This paper presents a ResNet-based model for automatic detection of sleep apnea syndrome (SAS) using raw electrocardiogram (ECG) signals. The model utilizes multi-layer convolutional and pooling structures to achieve efficient feature extraction, eliminating the need for additional feature engineering. To further enhance the model’s performance, a special attention mechanism was introduced, enabling the model to focus on key features in the signal, effectively improving the detection accuracy and robustness. Experimental results show that the model achieves excellent performance in terms of accuracy (93.84%), sensitivity (92.00%), specificity (94.97%), and F1 score (0.9193), significantly outperforming traditional network models. Compared with recent related studies, this model demonstrates significant advantages in the task of SAS detection, highlighting its potential for practical applications.